基于分类决策树的城市内部迁居人群通勤时间预测模型

针对迁居个体长距离、长时间的通勤现象，以南京迁居人群通勤出行多源数据为基础，综合考虑迁居人群个人家庭属性、迁居属性、建成环境属性及通勤属性，建立迁居人群通勤时间多元线性回归模型，分析各属性对通勤时间的影响. 结合多元线性回归和Logistic 回归确定决策树模型入选变量和通勤时间离散值，利用C4.5 算法建立迁居人群通勤时间决策树，以决策树结构揭示不同背景下迁居个体通勤时间模式特征. 结果显示，相较于个人家庭属性，职住同区、通勤出发时刻、住房类型、通勤距离、通勤方式及社区至主干道的便捷程度对迁居个体通勤时长的影响更为显著，侧面反映社区微观建成环境优化政策更利于提升迁居个体通勤效率.

Commute Time Pattern of Relocated Residents in City Based on Classification Decision Tree

Focusing on the long-distance and long-time commute of relocated residents, this study analyzed the impact of factors including personal and household characteristics, relocation attribute, built environment and commute attribute on commute time of relocated commuters by using multivariate linear regression model. After the identification of significant variables and the discretized value by linear regression model and Logistic regression model, a classification decision tree was developed to interpret the commute time pattern of relocated commuters in different scenarios by using the C4.5 algorithm. The result indicates that same zone of job-housing, departure time, house type, job-housing distance, commute mode, and accessibility of community to arterials are detrimental factors of commute time compared to personal and household characteristics, reflecting that policies that optimize the built environment in micro level contribute more in improving commute efficiency for relocated residents.